Re: [PATCH] tests: fix autostash

From: Ramkumar Ramachandra <artagnon@gmail.com>
Date: Jun 8, 2013, 13:19 UTC
Message-ID: <CALkWK0=qVXL=DtJzYRGFLQQh3_-cGC_t_zpaiTL9KVKN4--mVg@mail.gmail.com>
In-Reply-To: <CAMP44s3rYXOmGWR3JM8xy8ON4pqDRhYWVMQqOAyODVr5myq+yg@mail.gmail.com>

Felipe Contreras wrote:

Show 5 quoted lines

> Yes you do. The rest of the tests expect that the previous rebase has
> been aborted.
>
> In fact, all the tests depend on the previous test finishing
> correctly, which is not the way tests should be written.

How else am I supposed to write them? If there is a stale state from the previous test, there isn't too much I can do. Or should I be cleaning up state at the beginning of each test, instead of at the end?

> Doing 'rm -rf $dotest' is even worst than 'git rebase --abort',
> because it relies on the implementation of 'git rebase', which might
> need to remove more files than $dotest.

Huh?  Tests aren't allowed to rely on how a command is implemented?

  $ git grep test_path t

Ofcourse they're implementation details. Even in this very test, I check $dotest/autostash plenty of times.

Have you read rr/rebase-autostash? The whole idea is to inject $dotest/autostash and teach various scripts about how their assumptions about $dotest have changed.

> This wouldn't be a problem if the tests were implemented correctly,
> but they are not, so 'git rebase --abort' is the only sane option.

Then show me how to do it correctly.